Skip to main content Skip to secondary navigation
Main content start

A brain-inspired model can help predict COVID’s spread and recovery

A mechanical engineer adapts a computational model of the brain to create a tool that can help policymakers assess how various interventions might affect the course of COVID-19.

The web of neurons that make up the brain is an apt metaphor for a network of cities — only the travelers are people, not electrical signals. | Unsplash/Tomas Williams

The web of neurons that make up the brain is an apt metaphor for a network of cities — only the travelers are people, not electrical signals. | Unsplash/Tomas Williams

Ellen Kuhl is a mechanical engineer by training who has become one of the foremost designers of computational models that simulate living structures, like the heart or the brain.

In recent weeks, she quickly adapted one of her models to create a tool that can help policymakers assess how various interventions might affect the course of COVID-19 and how best to manage the recovery when the time comes. Learn more here.

You’re a mechanical engineer; how did you come to model a pandemic?

Only a few weeks ago we were working on modeling how diseases spread across the brain. And the brain is usually modeled as a network of connections. Now we adapted this network model as a model for air travel between the 50 states. The web of neurons that make up the brain is an apt metaphor for a network of cities — only the travelers are people, not electrical signals. Some of those travelers are infectious and will infect others in new locations.

The brain, of course, has very different dynamics than a population of people, so we replaced the disease model with data from the epidemiology of COVID-19, but the main infrastructure for the model was already coded. We only needed to adjust it slightly and change the parameters to predict the pattern of COVID-19 spread.

How does the model work?

We’ve created a computer model that represents the United States as a network with 50 nodes, one for each state. Each node has three parameters — let’s call them “control knobs” — one to account for the initial attention paid to the outbreak, one to dial-in how drastically a state is responding to the pandemic, and one to account for the number of people traveling in and out of the state.

For that, we are using real airline travel data. We hope to add other modes of travel later but the data on those is scarce. Air travel is probably the most relevant method of transportation between states like California and New York, for instance.

And, because we’re only at the early stages of the outbreak in the United States, we also try to learn from other countries. Like the average time it takes to show symptoms or to recover, which we can learn from countries like China that have already gone through a complete cycle of rise, peak and decay of infections. What’s most exciting is that we now adjust the model daily, in real time.

As we speak on April 6, it is day 75 of the U.S. outbreak. Our model estimates the national peak will be on day 110, in early May, but that’s if all things remain status quo. Additional travel restrictions or premature lifting constraints will affect that prediction. That’s why we monitor and update the model in real time.

What can your model tell us that others don’t?

I think people grasp the whole “flattening the curve” concept. But there are still a lot of questions, not just about the pandemic, but about the return to normal, that our model can address. What happens if we lock everybody in? How do we begin to release people from the shelter-in-place? How do we prevent the disease from starting again? You can get very granular and ask, for instance, “What happens if we send little kids back to school and their 30- to 40-year-old parents to work?” Various what-if scenarios like that will be very helpful for decision-makers.

 

Related Departments